2.4. Statistical analysis
All values of pollen density, RPF, and cross-pollination were log(x + 1) transformed to improve the normality of errors. To detectwhether Sorghum barrier reduces gene flow from GM maize to con-ventional maize, a general linear mixed effect model was fitted (Y∼Sorghum + (1|direction/distance)) using restricted maximum likeli-hood (REML), with Sorghum barrier as the fixed effect and directionand distance as random effects.To calculate the trends of pollen flow and cross-pollination rateover distances at each direction in the open and Sorghum experi-ments, two models were employed. Pollen number was ln (x + 1)transformed. Cross-pollination rate was square root transformed.First, a power trend line (y∼ a × exp(−b × x)) was fitted using thenonlinear least-squares estimates. Second, the distance was trans-formed with natural logarithm (ln) and a linear model (y∼ a + b × x)was employed to fit the trends. Diffslope (simba package in R) wasemployed to test the difference in slopes and intercepts of open andSorghum linear regression lines at each direction. The maximumthreshold distances at which no pollen load or cross-pollinationrate is equal to or lower than a threshold value of 0.9% (EuropeanUnion labeling threshold, EC 2003) at each direction were predictedusing the two models respectively. Paired students’ t-test was usedto test whether Sorghum barrier have effect on the observed andpredicted maximum distance of pollen flow and cross-pollinationrate. Statistical analysis was performed using R software with thepackages “lme4” and “simba” (R Development Core Team, 2012).